Multidimensional dietary patterns and cardiometabolic comorbidity in adults with anxiety disorders: evidence from NHANES and linked dietary databases using interpretable machine learning

Anxiety disorders frequently co-occur with cardiometabolic conditions, yet little is known about how everyday dietary exposures shape cardiovascular vulnerability within this population. Existing studies typically focus on isolated nutrients or single dietary indices, overlooking the broader food environment in which dietary choices are made. This study examined how multiple dimensions of diet including micronutrient intake, dietary patterns and specific food items relate to hypertension–coronary heart disease (CHD) comorbidity among adults with anxiety disorders. We analyzed data from seven NHANES cycles (2005–2018) to identify adults with anxiety disorders and assess the presence of hypertension–CHD comorbidity. Dietary exposures were evaluated using 24-hour recalls and aggregated across three complementary domains linked to NHANES: core micronutrients, MyPyramid Equivalents for food groups, and WWEIA for individual foods. Multicollinearity was assessed through correlation analysis, followed by feature selection using the Boruta algorithm. Class imbalance was addressed with the Synthetic Minority Over-sampling Technique (SMOTE). Six machine learning models were developed and compared. Feature-level associations between dietary micronutrients, daily dietary intake, and food consumption and the risk of hypertension–CHD comorbidity were interpreted using SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME). Model stability was further evaluated through time-series analyses. Across analytic approaches, several consistent dietary signatures emerged. Higher intakes of vitamin E, polyunsaturated fatty acids, folate, calcium, protein, and total energy were associated with lower odds of comorbidity, whereas zinc and iron intake showed positive associations. Dietary patterns rich in protein foods, poultry, seafood, and vegetables were inversely related to comorbidity, while starchy vegetables and cured meats were associated with increased risk. At the food level, bananas, coffee, cream substitutes, and tomato-based products were linked to lower risk, whereas yeast breads, margarine, cakes, and pies were linked to higher risk. Machine-learning models reinforced these findings and highlighted the cumulative influence of food choices rather than single nutrients. In adults with anxiety disorders, we identified clear and consistent dietary patterns associated with the coexistence of hypertension and coronary heart disease. By combining micronutrient intake, dietary patterns, and food-level data with interpretable machine-learning approaches, this study demonstrates that diet may contribute to cardiometabolic vulnerability in this understudied population. These findings highlight diet as a modifiable factor that could be incorporated into broader prevention efforts and support the value of routine dietary assessment in both psychiatric and general healthcare settings.